QUARTERLY JOURNAL OF ECONOMICS

Transcription

1 THE QUARTERLY JOURNAL OF ECONOMICS Vol. CXVIII February 2003 Issue 1 INCOME INEQUALITY IN THE UNITED STATES, * THOMAS PIKETTY AND EMMANUEL SAEZ This paper presents new homogeneous series on top shares of income and wages from 1913 to 1998 in the United States using individual tax returns data. Top income and wages shares display a U-shaped pattern over the century. Our series suggest that the large shocks that capital owners experienced during the Great Depression and World War II have had a permanent effect on top capital incomes. We argue that steep progressive income and estate taxation may have prevented large fortunes from fully recovering from these shocks. Top wage shares were at before World War II, dropped precipitously during the war, and did not start to recover before the late 1960s but are now higher than before World War II. As a result, the working rich have replaced the rentiers at the top of the income distribution. I. INTRODUCTION According to Kuznets in uential hypothesis, income inequality should follow an inverse-u shape along the development process, rst rising with industrialization and then declining, as more and more workers join the high-productivity sectors of the economy [Kuznets 1955]. Today, the Kuznets curve is widely held to have doubled back on itself, especially in the United States, with the period of falling inequality observed during the rst half * We thank Anthony Atkinson, Lawrence Katz, and two anonymous referees for their very helpful and detailed comments. We have also bene ted from comments and discussions with Daron Acemoglu, Philippe Aghion, Alberto Alesina, David Autor, Abhijit Banerjee, Francesco Caselli, Dora Costa, David Cutler, Esther Du o, Daniel Feenberg, William Gale, Claudia Goldin, Alan Krueger, Howard Rosenthal, and numerous seminar participants. We acknowledge nancial support from the MacArthur foundation. All our series are available in machine readable format in an electronic appendix of the working paper version at by the President and Fellows of Harvard College and the Massachusetts Institute of Technology. The Quarterly Journal of Economics, February

2 2 QUARTERLY JOURNAL OF ECONOMICS of the twentieth century being succeeded by a very sharp reversal of the trend since the 1970s. This does not, however, imply that Kuznets hypothesis is no longer of interest. One could indeed argue that what has been happening since the 1970s is just a remake of the previous inverse-u curve: a new industrial revolution has taken place, thereby leading to increasing inequality, and inequality will decline again at some point, as more and more workers bene t from the innovations. To cast light on this central issue, we build new homogeneous series on top shares of pretax income and wages in the United States covering the 1913 to 1998 period. These new series are based primarily on tax returns data published annually by the Internal Revenue Service (IRS) since the income tax was instituted in 1913, as well as on the large micro- les of tax returns released by the IRS since First, we have constructed annual series of shares of total income accruing to various upper income groups fractiles within the top decile of the income distribution. For each of these fractiles we also present the shares of each source of income such as wages, business income, and capital income. Kuznets [1953] did produce a number of top income shares series covering the 1913 to 1948 period, but tended to underestimate top income shares, and the highest group analyzed by Kuznets is the top percentile. 1 Most importantly, nobody has attempted to estimate, as we do here, homogeneous series covering the entire century. 2 Second, we have constructed annual 1927 to 1998 series of top shares of salaries for the top fractiles of the wage income distribution, based on tax returns tabulations by size of salaries compiled by the IRS since To our knowledge, this is the rst time that a homogeneous annual series of top wage shares starting before the 1950s for the United States has been produced. 3 Finally, in order to complete our analysis of top capital income earners, we have also used estate tax returns tabulations to construct quasi-annual series (1916 to 1997) of top estates. 1. Analyzing smaller groups within the top percentile is critical because capital income is extremely concentrated. 2. Feenberg and Poterba [1993, 2000] have constructed top income share series covering the period, but their series are not homogeneous with those of Kuznets. Moreover, they provide income shares series only for the top 0.5 percent, and not for other fractiles. 3. Previous studies on wage inequality before 1945 in the United States rely mostly on occupational pay ratios [Williamson and Lindert 1980; Goldin and Margo 1992; Goldin and Katz 1999].

3 INCOME INEQUALITY IN THE UNITED STATES, Our estimated top shares series display a U-shape over the century and suggest that a pure Kuznets mechanism cannot fully account for the facts. We nd that top capital incomes were severely hit by major shocks in the rst part of the century. The post-world War I depression and the Great Depression destroyed many businesses and thus signi cantly reduced top capital incomes. The wars generated large scal shocks, especially in the corporate sector that mechanically reduced distributions to stockholders. We argue that top capital incomes were never able to fully recover from these shocks, probably because of the dynamic effects of progressive taxation on capital accumulation and wealth inequality. We also show that top wage shares were at from the 1920s until 1940 and dropped precipitously during the war. Top wage shares have started to recover from the World War II shock in the late 1960s, and they are now higher than before World War II. Thus, the increase in top income shares in the last three decades is the direct consequence of the surge in top wages. As a result, the composition of income in the top income groups has shifted dramatically over the century: the working rich have now replaced the coupon-clipping rentiers. We argue that both the downturn and the upturn of top wage shares seem too sudden to be accounted for by technical change alone. Our series suggest that other factors, such as changes in labor market institutions, scal policy, or more generally social norms regarding pay inequality may have played important roles in the determination of the wage structure. Although our proposed interpretation for the observed trends seems plausible to us, we stress that we cannot prove that progressive taxation and social norms have indeed played the role we attribute to them. In our view, the primary contribution of this paper is to provide new series on income and wage inequality. One additional motivation for constructing long series is to be able to separate the trends in inequality that are the consequence of real economic change from those that are due to scal manipulation. The issue of scal manipulation has recently received much attention. Studies analyzing the effects of the Tax Reform Act of 1986 (TRA86) have emphasized that a large part of the response observable in tax returns was due to income shifting between the corporate sector and the individual sector [Slemrod 1996; Gordon and Slemrod 2000]. We do not deny that scal manipulation can have substantial short-run effects, but we argue that most long-run inequality trends are the consequence of

4 4 QUARTERLY JOURNAL OF ECONOMICS real economic change, and that a short-run perspective might lead to attribute improperly some of these trends to scal manipulation. The paper is organized as follows. Section II describes our data sources and outlines our estimation methods. In Section III we present and analyze the trends in top income shares, with particular attention to the issue of top capital incomes. Section IV focuses on trends in top wages shares. Section V offers concluding comments and compares our U. S. ndings with comparable series recently constructed for France by Piketty [2001a, 2001b] and for the United Kingdom by Atkinson [2001]. All series and complete technical details about our methodology are gathered in appendices of the working paper version of the paper [Piketty and Saez 2001]. II. DATA AND METHODOLOGY Our estimations rely on tax returns statistics compiled annually by the Internal Revenue Service since the beginning of the modern U. S. income tax in Before 1944, because of large exemptions levels, only a small fraction of individuals had to le tax returns and therefore, by necessity, we must restrict our analysis to the top decile of the income distribution. 4 Because our data are based on tax returns, they do not provide information on the distribution of individual incomes within a tax unit. As a result, all our series are for tax units and not individuals. 5 A tax unit is de ned as a married couple living together (with dependents) or a single adult (with dependents), as in the current tax law. The average number of individuals per tax unit decreased over the century but this decrease was roughly uniform across income groups. Therefore, if income were evenly allocated to individuals within tax units, 6 the time series pattern of top shares based on individuals should be very similar to that based on tax units. 4. From 1913 to 1916, because of higher exemption levels, we can provide estimates only within the top percentile. 5. Kuznets [1953] nevertheless decided to estimate series based on individuals not tax units. We explain in Piketty and Saez [2001] why his method produced a downward bias in the levels (though not in the pattern) of top shares. 6. Obviously, income is not earned evenly across individuals within tax units, and, because of increasing female labor force participation, the share of income earned by the primary earner has certainly declined over the century. Therefore, inequality series based on income earned at the individual level would be different. Our tax returns statistics are mute on this issue. We come back to that point when we present our wage estimates.

5 INCOME INEQUALITY IN THE UNITED STATES, TABLE I THRESHOLDS AND AVERAGE INCOMES IN TOP GROUPS WITHIN THE TOP DECILE IN 1998 Thresholds (1) Income level (2) Fractiles (3) Number of tax units (4) Average income (5) Full Population 130,945,000 $38,740 P90 $81,700 P ,550,000 $94,000 P95 $107,400 P ,240,000 $143,000 P99 $230,200 P ,000 $267,000 P99.5 $316,100 P ,000 $494,000 P99.9 $790,400 P ,900 $1,490,000 P99.99 $3,620,500 P ,100 $9,970,000 Computations are based on income tax returns statistics (see Piketty and Saez [2001], Appendix A). Income is de ned as gross income excluding capital gains and before individual taxes. Amounts are expressed in 1998 dollars. Source: Table A0 and Table A4, row 1998 in Piketty and Saez [2001]. Tax units within the top decile form a very heterogeneous group, from the high middle class families deriving most of their income from wages to the super-rich living off large fortunes. More precisely, we will see that the composition of income varies substantially by income level within the top decile. Therefore, it is critical to divide the top decile into smaller fractiles. Following Piketty [2001a, 2001b], in addition to the top decile (denoted by P90 100), we have constructed series for a number of higher fractiles within the top decile: the top 5 percent (P95 100), the top 1 percent (P99 100), the top 0.5 percent (P ), the top 0.1 percent (P ), and the top 0.01 percent (P ). This also allows us to analyze the ve intermediate fractiles within the top decile: P90 95, P95 99, P , P , P Each fractile is de ned relative to the total number of potential tax units in the entire U. S. population. This number is computed using population and family census statistics [U. S. Department of Commerce, Bureau of Census 1975; Bureau of Census 1999] and should not be confused with the actual number of tax returns led. In order to get a more concrete sense of size of income by fractiles, Table I displays the thresholds, the average income level in each fractile, along with the number of tax units in each fractile all for We use a gross income de nition including all income items reported on tax returns and before all deductions: salaries and wages, small business and farm income, partnership and du-

6 6 QUARTERLY JOURNAL OF ECONOMICS ciary income, dividends, interest, rents, royalties, and other small items reported as other income. Realized capital gains are not an annual ow of income (in general, capital gains are realized by individuals in a lumpy way) and form a very volatile component of income with large aggregate variations from year to year depending on stock price variations. Therefore, we focus mainly on series that exclude capital gains. 7 Income, according to our de nition, is computed before individual income taxes and individual payroll taxes but after employers payroll taxes and corporate income taxes. 8 The sources from which we obtained our data consist of tables displaying the number of tax returns, the amounts reported, and the income composition, for a large number of income brackets [U. S. Treasury Department, Internal Revenue Service, ]. As the top tail of the income distribution is very well approximated by a Pareto distribution, we use simple parametric interpolation methods to estimate the thresholds and average income levels for each of our fractiles. We then estimate shares of income by dividing the income amounts accruing to each fractile by total personal income computed from National Income Accounts [Kuznets 1941, 1945; U. S. Department of Commerce 2000]. 9 Using the published information on composition of income by brackets and a simple linear interpolation method, we decompose the amount of income for each fractile into ve components: salaries and wages, dividends, interest income, rents and royalties, and business income. We use the same methodology to compute top wage shares using published tables classifying tax returns by size of salaries and wages. In this case, fractiles are de ned relative to the total number of tax units with positive wages and salaries estimated as the number of part-time and full workers from National Income Accounts [U. S. Department of Commerce 2000] less the number 7. In order to assess the sensitivity of our results to the treatment of capital gains, we present additional series including capital gains (see below). Details on the methodology and complete series are presented in the appendix to Piketty and Saez [2001]. 8. Computing series after individual income taxes is beyond the scope of the present paper but is a necessary step to analyze the redistributive power of the income tax over time, as well as behavioral responses to individual income taxation. 9. This methodology using tax returns to compute the level of top incomes, and using national accounts to compute the total income denominator is standard in historical studies of income inequality. Kuznets [1953], for instance, adopted this method.

7 INCOME INEQUALITY IN THE UNITED STATES, of wives who are employees (estimated from U. S. Department of Commerce, Bureau of Census [1975] and Bureau of Census [1999]). The sum of total wages in the economy used to compute shares is also obtained from National Income Accounts [U. S. Department of Commerce 2000]. The published IRS data vary from year to year, and there are numerous changes in tax law between 1913 and To construct homogeneous series, we make a number of adjustments and corrections. Individual tax returns micro- les are available since They allow us to do exact computations of all our statistics for that period and to check the validity of our adjustments. Kuznets [1953] was not able to use micro- les to assess possible biases in his estimates due to his methodological assumptions. 12 Our method differs from the recent important studies by Feenberg and Poterba [1993, 2000] who derive series of the income share of the top 0.5 percent 13 for 1951 to They use total income reported on tax returns as their denominator and the total adult population as their base to obtain the number of tax units corresponding to the top fractiles. 14 Their method is simpler than ours but cannot be used for years before 1945 when a small fraction of the population led tax returns. III. TOP INCOME SHARES AND COMPOSITION III. A. Trends in Top Income Shares The basic series of top income shares are presented in Table II. Figure I shows that the income share of the top decile of tax units from 1917 to 1998 is U-shaped. The share of the top decile uctuated around 40 to 45 percent during the interwar period. It declined substantially to about 30 percent during World War II and then remained stable at 31 to 32 percent until the 1970s when it increased again. By the mid-1990s the share had crossed the 40 percent level and is now at a level close to the prewar level, 10. The most important example is the treatment of capital gains and the percentage of these gains that are included in the statistics tables. 11. These data are known as the Individual Tax Model les. They contain about 100,000 returns per year and largely oversample high incomes, providing a very precise picture of top reported incomes. 12. In particular, Kuznet s treatment of capital gains produces a downward bias in the level of his top shares. 13. They also present incomplete series for the top 1 percent. 14. This method is not fully satisfying for a long-run study as the average number of adults per tax unit has decreased signi cantly since World War II.

8 8 QUARTERLY JOURNAL OF ECONOMICS TABLE II TOP INCOME SHARES, In ation CPI (p(1998)/p(n)) # tax units (thousands) Average income (1998 $) Top income shares (excluding capital gains) Capital gains included P P90 95 P95 99 P P P P Share only P Rank and share P Year (0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) ,701 12, ,513 11, ,154 11, ,790 12, ,387 13, ,451 12, ,052 12, ,909 10, ,835 9, ,543 10, ,409 12, ,384 11, ,190 12, ,940 12, ,723 12, ,445 12, ,085 13, ,750 11, ,462 10, ,117 9, ,757 8, ,430 9, ,147 10, ,844 11, ,539 12, ,342 11,

9 INCOME INEQUALITY IN THE UNITED STATES, ,181 11, ,115 12, ,392 14, ,736 16, ,250 19, ,656 20, ,997 20, ,297 19, ,118 18, ,825 18, ,537 18, ,446 19, ,060 20, ,684 21, ,273 21, ,928 21, ,589 23, ,257 24, ,947 24, ,546 23, ,144 25, ,681 25, ,997 25, ,254 26, ,464 27, ,660 28, ,772 29, ,831 31, ,856 31, ,826 33, ,793 33, ,924 33, ,849 33, ,670 34, ,442 35, ,228 34, ,127 32,

10 10 QUARTERLY JOURNAL OF ECONOMICS TABLE II Continued In ation CPI (p(1998)/p(n)) # tax units (thousands) Average income (1998 $) Top income shares (excluding capital gains) Capital gains included P P90 95 P95 99 P P P P Share only P Rank and share P Year (0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) ,048 33, ,076 33, ,213 34, ,457 34, ,625 32, ,432 31, ,250 31, ,067 31, ,871 32, ,736 32, ,684 33, ,640 34, ,656 35, ,759 35, ,055 35, ,453 34, ,944 34, ,378 33, ,716 34, ,023 34, ,625 35, ,301 37, ,945 38, Full details are in Appendix A of Piketty and Saez [2001]. Number of tax units is estimated. Total income is estimated from National Income Accounts. Top shares are obtained from income tax returns statistics and Pareto interpolation. Top shares in columns (3) to (9) are based on individual gross income before individual taxes and excluding capital gains. Column (10) displays top percentile (based on income excluding capital gains) share where capital gains are added back into the share computation. Column (11) displays top percentile (based on income including capital gains) share where capital gains are added back into the share computation. Source: Tables A0 and A1 in Piketty and Saez [2001].

11 INCOME INEQUALITY IN THE UNITED STATES, FIGURE I The Top Decile Income Share, Source: Table II, column P although a bit lower. Therefore, the evidence suggests that the twentieth century decline in inequality took place in a very speci c and brief time interval. Such an abrupt decline cannot easily be reconciled with a Kuznets-type process. The smooth increase in inequality in the last three decades is more consistent with slow underlying changes in the demand and supply of factors, even though it should be noted that a signi cant part of the gain is concentrated in 1987 and 1988 just after the Tax Reform Act of 1986 which sharply cut the top marginal income tax rates (we will return to this issue). Looking at the bottom fractiles within the top decile (P90 95 and P95 99) in Figure II reveals new evidence. These fractiles account for a relatively small fraction of the total uctuation of the top decile income share. The drop in the shares of fractiles P90 95 and P95 99 during World War II is less extreme than that for the top decile as a whole, and they start recovering from the World War II shock directly after the war. These shares do not increase much during the 1980s and 1990s (the P90 95 share was fairly stable, and the P95 99 share increased by about 2 percentage points while the top decile share increased by about 10 percentage points).

12 12 QUARTERLY JOURNAL OF ECONOMICS FIGURE II The Income Shares of P90 95, P95 99, and P99 100, Source: Table II, columns P90 95, P95 99, and P In contrast to P90 95 and P95 99, the top percentile (P in Figure II) underwent enormous uctuations over the twentieth century. The share of total income received by the top 1 percent was about 18 percent before World War I, but only about 8 percent from the late 1950s to the 1970s. The top percentile share declined during World War I and the postwar depression (1916 to 1920), recovered during the 1920s boom, and declined again during the Great Depression (1929 to 1932, and 1936 to 1938) and World War II. This highly speci c timing for the pattern of top incomes, composed primarily of capital income (see below), strongly suggests that shocks to capital owners between 1914 and 1945 (depression and wars) played a key role. The depressions of the interwar period were far more profound in their effects than the post-world War II recessions. As a result, it is not surprising that the uctuations in top shares were far wider during the interwar period than in the decades after the war The fact that top shares are very smooth after 1945 and bumpy before is therefore not an artifact of an increase in the accuracy of the data (in fact, the data are more detailed before World War II than after), but re ects real changes in the economic conditions.

13 INCOME INEQUALITY IN THE UNITED STATES, Figure II shows that the uctuation of shares for P90 95 and P95 99 is exactly opposite to the uctuation for P over the business cycle from 1917 to As shown below, the P90 95 and P95 99 incomes are mostly composed of wage income, while the P incomes are mostly composed of capital income. During the large downturns of the interwar period, capital income sharply fell while wages (especially for those near the top), which are generally rigid nominally, improved in relative terms. On the other hand, during the booms ( ) and the recovery ( ), capital income increased quickly, but as prices rose, top wages lost in relative terms. 16 The negative effect of the wars on top incomes is due in part to the large tax increases enacted to nance them. During both wars, the corporate income tax (as well as the individual income tax) was drastically increased and this mechanically reduced the distributions to stockholders. 17 National Income Accounts show that during World War II, corporate pro ts surged, but dividend distributions stagnated mostly because of the increase in the corporate tax (that increased from less than 20 percent to over 50 percent) but also because retained earnings increased sharply. 18 The decline in top incomes during the rst part of the century is even more pronounced for higher fractiles within the top percentile, groups that could be expected to rely more heavily on capital income. As depicted in Figure III, the income share of the top 0.01 percent underwent huge uctuations during the century. In 1915 the top 0.01 percent earned 400 times more than the average; in 1970 the average top 0.01 percent income was only 50 times the average; in 1998 they earned about 250 times the average income. Our long-term series place the TRA86 episode in a longer term perspective. Feenberg and Poterba [1993, 2000], looking at the top 0.5 percent income shares series ending in 1992 (respectively, 1995), argued that the surge after TRA86 appeared permanent. However, completing the series up to 1998 shows that the signi cant increase in the top marginal tax rate, from 31 to 16. Piketty [2001a, 2001b] shows that exactly the same phenomenon is taking place in France during the same period. 17. During World War I, top income tax rates reached modern levels above 60 percent in less than two years. As was forcefully argued at that time by Mellon [1924], it is conceivable that large incomes found temporary ways to avoid taxation at a time when the administration of the Internal Revenue Service was still in its infancy. 18. Computing top shares for incomes before corporate taxes by imputing corporate pro ts corresponding to dividends received is an important task left for future research (see Goldsmith et al. [1954] and Cartter [1954] for such an attempt around the World War II period).

14 14 QUARTERLY JOURNAL OF ECONOMICS FIGURE III The Top 0.01 Percent Income Share, Source: Table II, column P percent, enacted in 1993 on did not prevent top shares from increasing sharply. 19 From that perspective, looking at Figures II and III, the average increase in top shares from 1985 to 1994 is not signi cantly higher than the increase from 1994 to 1998 or from 1978 to As a result, it is possible to argue that TRA86 produced no permanent surge in top income shares, but only a transitory blip. The analysis of top wage shares in Section IV will reinforce this interpretation. In any case, the pattern of top income shares cannot be explained fully by the pattern of top income tax rates. III. B. The Secular Decline of Top Capital Incomes To demonstrate more conclusively that shocks to capital income were responsible for the large decline of top shares in the rst part of the century, we look at the composition of income within the top fractiles. Table III reports the composition of income in top groups for various years from 1916 and Figure IV displays the composition of income for each fractile in 19. Slemrod and Bakija [2000] pointed out that top incomes have surged in recent years. They note that tax payments by taxpayers with AGI above $200,000 increased signi cantly from 1995 to 1997.

15 INCOME INEQUALITY IN THE UNITED STATES, TABLE III INCOME COMPOSITION BY SIZE OF TOTAL INCOME, P90 95 P95 99 P P P P Wage Entrep. K K inc. gains Wage Entrep. K K inc. gains Wage Entrep. K K inc. gains Wage Entrep. K K inc. gains Wage Entrep. K K inc. gains Wage Entrep. K K inc. gains Fractiles are de ned by size of total income (excluding capital gains). For each fractile, the rst three columns (summing to 100 percent) give the percentage of wage income (wages and salaries), entrepreneurial income (self-employment income, partnership income, and small business income), and capital income (dividends, interest, and rents) in total income (excluding capital gains). The fourth column displays the percentage of income (de ned including capital gains) obtained by each of these fractiles (de ned again by total income excluding capital gains) from capital gains. Details on methodology are presented in Appendix A of Piketty and Saez [2001]. Source: Computations are based on Individual Income Statistics and are reported in Tables A7 and A8, various rows, in Piketty and Saez [2001].

16 16 QUARTERLY JOURNAL OF ECONOMICS FIGURE IV Income Composition of Top Groups within the Top Decile in 1929 and 1998 Capital income does not include capital gains. Source: Table III, rows 1929 and 1998.

17 INCOME INEQUALITY IN THE UNITED STATES, (Panel A) and 1998 (Panel B). As expected, Panel A shows that the share of wage income is a declining function of income and that the share of capital income (dividends, interest, rents, and royalties) is an increasing function of income. The share of entrepreneurial income (self-employment, small businesses, and partnerships) is fairly at. Thus, individuals in fractiles P90 95 and P95 99 rely mostly on labor income (capital income is less than 25 percent for these groups), while individuals in the top percentile derive most of their income in the form of capital income. Complete series in Piketty and Saez [2001] show that the sharply increasing pattern of capital income is entirely due to dividends. This evidence con rms that the very large decrease of top incomes observed during the 1914 to 1945 period was to a large extent a capital income phenomenon. One might also be tempted to interpret the large upturn in top income shares observed since the 1970s as a revival of very high capital incomes, but this is not the case. As shown in Panel B, the income composition pattern has changed drastically between 1929 and In 1998 the share of wage income has increased signi cantly for all top groups. Even at the very top, wage income and entrepreneurial income form the vast majority of income. The share of capital income remains small (less than 25 percent) even for the highest incomes. Therefore, the composition of high incomes at the end of the century is very different from those earlier in the century. Before World War II, the richest Americans were overwhelmingly rentiers deriving most of their income from wealth holdings (mainly in the form of dividends). Occupation data by income bracket were published by the IRS in These data show that, at the very top, the vast majority of taxpayers reported themselves as Capitalists: Investors and Speculators, while a small fraction reported themselves as salaried workers (see Piketty and Saez [2001], Table 3 for details). In contrast, in 1998 more than half of the very top taxpayers derive the major part of their income in the form of wages and salaries. Thus, today, the working rich celebrated by Forbes magazine have overtaken the coupon-clipping rentiers. The dramatic evolution of the composition of top incomes appears robust and independent from the erratic evolution of capital gains excluded in Figures I to IV. The last two columns of Table II display the top 1 percent share including realized capital gains. In column (10), in order to get around the lumpiness of realizations, individuals are ranked by income excluding capital

18 18 QUARTERLY JOURNAL OF ECONOMICS FIGURE V The Capital Income Share in the Top 0.5 Percent, Series display the share of capital income (excluding capital gains) and dividends in total income (excluding capital gains) for the top 0.5 percent income quantile. Source: Authors computations are based on income tax returns statistics (series reported in Piketty and Saez [2001], Table A7, column P ). gains, but capital gains are added back into income to compute shares. In column (11) individuals are ranked by income including capital gains, and capital gains are added back into income to compute shares. These additional series show that including capital gains does not modify our main conclusion that very top income shares dropped enormously during the period before increasing steadily in the last three decades. 20 The decline of the capital income share is a very long-term phenomenon and is not limited to a few years and a few thousand tax units. Figure V shows a gradual secular decline of the share of capital income (again excluding capital gains realizations) and dividends in the top 0.5 percent fractile from the 1920s to the 1990s: capital income was about 55 percent of total income in the 1920s, 35 percent in the 1950s 1960s, and 15 percent in the 20. It is interesting, however, to note that during the 1960s, when dividends were strongly tax disadvantaged relative to capital gains, capital gains do seem to represent a larger share in top incomes than during other periods such as the 1920s or late 1990s that also witnessed large increases in stock prices.

19 INCOME INEQUALITY IN THE UNITED STATES, s. Sharp declines occurred during World War I, the Great Depression, and World War II. Capital income recovered only partially from these shocks in the late 1940s and started a steady decline in the mid-1960s. This secular decline is entirely due to dividends: the share of interest, rent, and royalties has been roughly at while the dividend share has dropped from about 40 percent in the 1920s, to about 25 percent in the 1950s and 1960s, to less than 10 percent in the 1990s. 21 Most importantly, the secular decline of top capital incomes is due to a decreased concentration of capital income rather than a decline in the share of capital income in the economy as a whole. As displayed in Figure VI, the National Income Accounts series show that the aggregate capital income share has not declined over the century. As is well-known, factor shares in the corporate sector have been fairly at in the long run with the labor share around percent, and the capital share around percent (Panel A). The share of capital income in aggregate personal income is about 20 percent both in the 1920s and in the 1990s (Panel B). Similarly, the share of dividends was around 5 percent in the late 1990s and only slightly higher (about 6 7 percent) before the Great Depression. This secular decline is very small compared with the enormous fall of top capital incomes. 22 Contrary to a widely held view, dividends as a whole are still alive and well. 23 It should be noted, however, that the ratio of total dividends reported on individual tax returns to personal dividends in National Accounts has declined continuously over the period 1927 to 1995, starting from a level close to 90 percent in 1927, declining slowly to 60 percent in 1988, and dropping precipitously to less than 40 percent in This decline is due mostly to the growth of funded pension plans and retirement saving accounts through which individuals receive dividends that are never reported as dividends on income tax returns. For the highest income earners, 21. Tax statistics by size of dividends analyzed in Piketty and Saez [2001] con rm a drastic decline of top dividend incomes over the century. In 1998 dollars, top 0.1 percent dividends earners reported on average about $500,000 of dividends in 1927 but less than $240,000 in The share of dividends in personal income starts declining in 1940 because the corporate income tax increases sharply and permanently, mechanically reducing pro ts that can be distributed to stockholders. 23. As documented by Fama and French [2000], a growing fraction of rms never pay dividends (especially in the new technology industries, where rms often make no pro t at all), but the point is that total dividend payments continue to grow at the same rate as aggregate corporate pro ts.

20 FIGURE VI Capital Income in the Corporate and Personal Sector, Source: Authors computations are based on National Income and Product Accounts. Panel A from NIPA Table 1.16; consumption of xed capital and net interest have been included in the capital share. Panel B from NIPA Table 2.1; capital income includes dividends, interest, and rents.

21 INCOME INEQUALITY IN THE UNITED STATES, this additional source of dividends is likely to be very small relative to dividends directly reported on tax returns. Estate tax returns statistics (available since the beginning of the estate tax in 1916) are an alternative important source of data to analyze the evolution of large fortunes. 24 Lampman [1962] used these data to construct top 1 percent wealth shares for a few years between 1922 and 1956 using the estate multiplier method. We have constructed quasi-annual series of average levels (in 1998 dollars) of gross estates for various fractiles of decedents aged 25 and above (ranked by size of gross estate). Panel A in Figure VII displays the average level of gross estates for the top 0.01 percent of decedents from 1916 to 1997 (these are the largest 225 estates in 1997). Strikingly, the real value of the top estates in 1916 is about the same as in 1997, namely around $80 million, even though the GDP per capita grew by a factor of 3.5 during this period. Therefore, the biggest fortunes have in fact substantially declined in relative terms. 25 To emphasize this point, Panel B displays the evolution of average estates in lower fractiles. The average estate in P98 99 has grown by a factor 3 between 1916 and 1997, and the average estate in P has been multiplied by about 2.5. This evidence is consistent with our previous results on the decline in top capital incomes over the century. Popular accounts suggest that estate tax evasion is very important [Cooper 1979], but academics disagree about the extent of tax evasion [Poterba 2000]. Furthermore, our results would be invalidated only if the level of tax evasion had increased over time much more for the largest estates (top 0.01 percent) than for large estates. III. C. Proposed Interpretation: The Role of Progressive Taxation How can we explain the steep secular decline in capital income concentration? It is easy to understand how the macroeconomic shocks of the Great Depression and the scal shocks of World War I and World War II have had a negative impact on capital concentration. The dif cult question to answer is why large fortunes did not recover from these shocks. The most nat- 24. In particular, capital gains not realized before death are never reported on income tax returns, but are included in the value of assessed estates. 25. It is important to keep in mind that estate data re ect the wealth distribution of decedents and thus probably introduce a long lag relative to the current wealth distribution.

22 22 QUARTERLY JOURNAL OF ECONOMICS FIGURE VII Evolution of Estates (in real 1998 dollars), Source: Authors computations are based on estate tax returns statistics [Piketty and Saez 2001, Appendix C, Table C3]. Series report real value of gross estates before deductions (in 1998 dollars) for fractiles P (Panel A) and P98 99, P (Panel B) of decedents aged 25 and above.

23 INCOME INEQUALITY IN THE UNITED STATES, ural and realistic candidate for an explanation seems to be the creation and the development of the progressive income tax (and of the progressive estate tax and corporate income tax). The very large fortunes that generated the top 0.01 percent incomes observed at the beginning of the century were accumulated during the nineteenth century, at a time where progressive taxes hardly existed and capitalists could dispose of almost all their income to consume and to accumulate. 26 The scal situation faced by capitalists in the twentieth century to recover from the shocks incurred during the 1914 to 1945 period has been substantially different. Top tax rates were very high from the end of World War I to the early 1920s, and then continuously from 1932 to the mid-1980s. Moreover, the United States has imposed a sharply progressive estate tax since 1916, and a substantial corporate income tax ever since World War II. 27 These very high marginal rates applied to only a very small fraction of taxpayers, but created a substantial burden on the very top income groups (such as the top 0.1 percent and 0.01 percent) composed primarily of capital income. In contrast to progressive labor income taxation, which simply produces a level effect on earnings through labor supply responses, progressive taxation of capital income has cumulative or dynamic effects because it reduces the net return on wealth which generates tomorrow s wealth. It is dif cult to prove in a rigorous way that the dynamic effects of progressive taxation on capital accumulation and pretax income inequality have the right quantitative magnitude and account for the observed facts. One would need to know more about the savings rates of capitalists how their accumulation strategies have changed since The orders of magnitude do not seem unrealistic, especially if one assumes that the owners of large fortunes, whose pretax incomes were already severely hit by the prewar shocks, were not willing to reduce their consumption to very low levels. Piketty [2001a, 2001b] provides simple numerical simulations showing that for a xed saving rate, introducing substantial capital income taxation has a tremendous effect on the time needed to reconstitute large wealth holdings after negative shocks. Moreover, reduced savings in response to a reduction in the after-tax rate of return on wealth would accelerate the 26. During the nineteenth century, the only progressive tax was the property tax, but its level was low (see Brownlee [2000] for a detailed description). 27. From 1909 ( rst year the corporate tax was imposed) to the beginning of World War II, the corporate tax rate was low, except during World War I.

24 24 QUARTERLY JOURNAL OF ECONOMICS decrease in wealth inequality. Piketty [2001b] shows that in the classic dynastic model with in nite horizon, any positive capital income tax rate above a given high threshold of wealth will eventually eliminate all large wealth holdings without, however, affecting the total capital stock in the economy. We are not the rst to propose progressive taxation as an explanation for the decrease in top shares of income and wealth. Lampman [1962] did as well, and Kuznets [1955] explicitly mentioned this mechanism as well as the shocks incurred by capital owners during the 1913 to 1948 period, before presenting his inverted U-shaped curve theory based on technological change. Explanations pointing out that periods of technological revolutions such as the last part of the nineteenth century (industrial revolutions) or the end of the twentieth century (computer revolution) are more favorable to the making of fortunes than other periods might also be relevant. 28 Our results suggest that the decline in income tax progressivity since the 1980s and the projected repeal of the estate tax might again produce in a few decades levels of wealth concentration similar to those at the beginning of the century. IV. TOP WAGE SHARES Table IV displays top wage shares from 1927 to 1998 constructed using IRS tabulations by size of wages. There are three caveats to note about these long-term wage inequality series. First, self-employment income is not included in wages, and therefore our series focus only on wage income inequality. As self-employment income has been a decreasing share of labor income over the century, it is conceivable that the pool of wage and salary earners has substantially evolved over time, and that total labor income inequality series would differ from our wage inequality series. Second and related, large changes in the wage force due to the business cycle and wars might affect our series through compositional effects because we de ne the top fractiles relative to the total number of tax units with positive wage income. As can be seen in column (1) of Table II, the number of tax units with wages declined during the Great Depression due to 28. De Long [1998] also points out the potential role of antitrust law. According to De Long, antitrust law was enforced more loosely before 1929 and since 1980 than between 1929 and 1980.